Sistemas Vol. 6No. 3 pp. 196-203 “IPN, ISSN 1405-5546. Impreso en México Noisy Binary Texture Recognition Using the Coordinated Cluster Transform Reconocimiento de Texturas Binarias Ruidosas Usando 1a Transformada de Ciimulos Coordinados Evguenii Kurmyshev, Francisco Cuevas and Radl Sanchez Centro de Investigaciones en éptica, A.C. Apartado Postal 1-948 Leén Guanajuato, Mi E-mails: (key, fjeuevas, sanchez} @foton.cio.mx. Anticle rece Abstract In this paper a technique using the coordinated cluster representation (CCR) is examined for recognition of binary computer generated and natural texture images corrupted by additive noise. A normalized local property histogram of the CCRis used as @ unique feature vector, The ability of the descriptor to capture spatial statistical features of an image is exploited. The evaluation criteria is the recognition performance using a simple minimum distance classifier for recognition purposes, The experimental results indicate that the proposed technigueis efficient for recognition of textures deteriorated by high level additive noise. Textures under test run through periodic up to random ones. Keywords: Patter recognition, binary texture analysis, image representation, coordinated clusters. Resumen En este articulo se estudia una técnica, basada en la representacién de imégenes por ciimulos coordinadas (RICC), para el reconocimiento de imagenes binarias tanto de texturas naturales como aguellas generadas por computadara, las cuales fueron corrompidas por un ruido aditivo. El histograma normalizado de RIC es usado como vector tinico de caracteristicas de la imagen. explora la habilidad del descriptor de captar las caracteristicas estadisticas espaciales de una imagen. Como un criterio de iuacién usamos la eficiencia de reconocimiento usando un clasificador simple de distancia minima para el reconocimiento. Se ‘muestra que la técnica propuesta es eficiente para el reconocimiento de texturas deterioradas por el ruido aditivo, Se prueba la eficiencia del método en un rango amplio de texturas, sienda éstas desde puramente periédicas hasta completamente aleatorias. Palabras clave: Reconocitiento de patrones, andlisis de texturas binarias, representacion de imagenes, edirmulos coordinados, 196) ed on May 6, 2002: accepted on March 14, 2003 1 Introduction (One of the purposes of any pattern recognition technique is to carry out the classification of test images using a set of attributes extracted from pattern images. The problem is to establish a set of significant attributes that can be efficiently used in the classification. The set of features should obey the following requirements: they should be independent and the cardinality of the set should be minimal to facilitate computer manipulation, Feature extraction techniques can be classified into four major categories: statistical, model based, signal processing and structural (Tuceryan and Jain, 1993). After the attributes are chosen, textures can be classified using one of the well-known pattern classification methods (Duda ef al., 2001; Kahil and Bongard, 1970; Chen etal., 1996; Fukunga, 1990; Servin and Cuevas, 1993) Images of natural textures are difficult to classify or recognize due to the presence of statistic component in the intensity and/or color distribution of an image. In real experiments, one can find both full-periodic and random textures (Tuceryan and Jain, 1993; Haralick, 1979; Berry and Goutsias, 1989; Ohanian and Dubes, 1992). The well-known techniques of random, natural and artificial textures classification are mainly based on correlation moments and co-occurrence matrices, both approaches are time consuming (Heralick, 1979; Berry and Goutsias, 1989; Ohanian and Dubes, 1992; Elfadel and Picard, 1994; Soh and Tsatsoulis, 1999; Goon and Rolland, 199; Chetverikov, 1999). Some techniques are based on Markov random fields (Chellappa and Chatterjee, 1987) and others on window transforms (Turner, 1986; Azencott y Wang, 1997; Wang and He, 1990; Ojala et al.,1996; Valkealahti and Oja, 1998; Randen and Husoy, 1999)E. Kurmyshey, F. Cuevas, R. Sanchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform Recently a model for binary image representation called the coordinated cluster representation (CCR) was proposed in (Kurmyshey and Cervantes, 1996; Kurmyshev and Soto, 1996). In this model a binary image is characterized by a histogram of the occurrence of pattern units. The aim of this work is to evaluate the use of the histogram as a descriptor of textures. A simple minimum distance classifier using the unique attribute, a local property histogram generated by the coordinated cluster transform, is examined. We report the results when the CCR technique is applied to the recognition of a set of binary texture images degraded by noise. The CCR technique is based on binary images and, thus, they are of special interest to determine the limitations of the technique In addition, it is well known that the results obtained with simulated images can differ from that for real images where the texture can be irregular and a noise has to be estimated or supposed. In experiments with simulated images we have total control over texture and, in particular, over @ noise. Thus, experiments of this work have been implemented with binary simulated texture images on purpose. The paper is organized as follows. Section 2 describes briefly the coordinated cluster transform of binary images and its mathematical background, In Section 3 we outline an approach to binary texture image classification and recognition. The classification setup and experimental results for binary texture images are given in Section 4. Conclusions and directions of future work are given finally in Section 5. 2 Coordinated Cluster Transform In order to avoid some complicated mathematical notations, we describe coordinated cluster transform in an algorithmic way. Let S“=[sf,] be a matrix of binary image intensities m=1,2, (binary pir W=LxM and a Is) where [=12,..L and 2" is the binary image index. In order to calculate the CCR of a binary image S® one needs first (o establish a rectangular window of N = JxJ pixels (11 and J $M ) and then scan sequentially by means of this window all over the image S“ with one pixel step. Given ¢ S$, the coordinated cluster transformation generates the histogram of occurrence of pixel patterns that the scanning window has detected. The binary number of pixel configuration is used as a code for the configuration found by the window. Because pixels are binary units, the number ofall possible states of a window of the size N= x. is equal to 2”. This number determines the "length" of a primary histogram that can be reduced if one discards alll the states with zero of occurrence while scanning an image. Thus, the coordinated cluster transformation assigns to each binary a binary ima image S* a histogram H()(b) that represents the occurrence of window pattem, here « is the index of an image, the subscript (,./) indicates the size of the scanning 2" is the code number of window states. Itis clear that the total number of occurrences, that is the total area of the histogram, is equal to A=(L-141)x(M—J+1). The coordinated transform belongs to the kind of window transforms with no information preserved over the scanning window position, When a histogram Hj ;)(b) is normalized, itis considered as window and the variable b = 1,2, cluster a distribution function of occurrences Hine) Fin@)= 2 a) The distribution function can also be interpreted as an image spectrum in terms of the texture units; those were originally called coordinated clusters, in order to outline some statistical properties of the transform. The texture units of CCR are conceptually and mathematically different from those introduced by Wang and He (1990) and Ojala et al. (1996) where a binary code is used for the gray level pattern description. An example of 3%3 window patterns and their binary codes are shown in Figure I, where the binary coded decimal (BCD) corresponds to the index in the CCR histogram (and also in the distribution function). poten SeryCove BCD FF won EE wore: 89 Be erm 200 Figure 1: Window binary pattern and its binary coded decima as the index of CCR histogram, ‘The fundamental properties of the CCR were established in the two theorems (Kurmyshev and Cervantes, 1996; Kurmyshev and Soto, 1996) that are given here in a modified form, The first theorem establishes the structure of the CCR of periodic images. ‘Theorem 1. If a texel (primitive cell) of a transnational invariant binary image S has the size 7, Xt, pixels, x, in distribution one and t, in other direction, then any C fanction F@,)(6) takes nto more than T= 1, x7) non-zero values. If the CCR scanning window has the size equal to or larger than the texel size, 7, and J2r,, then Fi_,)(b) takes exactly T= 1, xr, non-zero values.E, Kurmyshey, F. Cuevas, R. Sanchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform ‘The second theorem establishes the relation betws 1; ,)(b)and spatial correlation moments of n-th order (n-th order statistics) of a binary image S“ second theorem, the According to the histogram Hf(6) contains all information about the n-point correlation moments of the image S* if and only if the separation vectors between 1 pixels fit into the scanning window, that is a distribution function Fi )(b) provides sufficient information about n= point joint probability functions. Theorem 2, Let S =[sf,] be the matrix ofa binary im: and F(j_)(6) be the CCR distribution function of the image. If max {lj} 1 C= 1 QoL) kS1x,J , then any autocorrelation function of k-th order and max{m,}< J and 5% (Lm)s@(U+1, mt). (E+ jms) rag Dos Cas" (+h, mrtm).s” lim Ww (+4 j.m+ my) can be uniquely reconstructed from Fif.)(b), where W=lxM is ue fn} Itis well recognized that the second and higher order joint probability density functions provide structural information abouta gray level image, being the correlation moments a set of significant attributes for the classification of images (Liu and Jernigan, 1990; ‘Thomson and Foster, 1997). On the other hand, there is a structural correspondence between a gray level image and its binarized analog. A binarized image keeps quite enough structural information about a primary gray level image to be classified. Thus, the CCR of a binary image, which preserves extremely well its structural information, is highly suitable for recognition and classification of gray level texture images having statistical properties in a wide range, from totally periodic to totally random textures included (Kurmyshev and Sanchez-Yéfiez, 2001). The CCR of each image can be implemented in a bitwise-oriented algorithm, allowing a fast computation of each number representing the local binary pattern. In order to preserve structural information of an image itis necessary to choose a CCR scanning window with a side approximately equal to or larger than the correlation distance between pixels. When a correlation distance is large, the scanning window has to be proportionally large too, Then the CCR histogram becomes rather large and complicated for computer management. One practical way to reconcile the two contradictory requirements (information preservation and the image size, L'= L~max {i,} M ~max 198 manageability of CCR) is to reduce the scale of images to be classified, 3 Texture Pattern Recognition In this work a set of noisy binary texture images was used to test the performance of the proposed recognition technique (see also Liu and Jernigan, 1990). We proceed in the following way. For a given set of binary texture images SP =[sf,} (p=1,2...P), we generate a set of noisy images {sr bt,)= ol pattern textures. In this work we use particularly a uniform additive noise. Several images are generated for each k-th noise level (SNR value). If Ais the difference between the two nearest noise evels, then the images s?(kc,)=[bp(&,a,)| for the sth noise level can be calculated as follows: satka,) ff if rand <1—kA [e.c,)} by adding @ noise t0 these Q) otherwise where a, indicates the @,, -sh seedling of noise to be added to 8’, k= 012, number generator function with uniform distribution in the range (0,1). When k = 0, the image generated by the Eq, 2) coincides with the original pattern texture, $”(0,a,,)= 5” Afier the set B is generated, the coordinated cluster transform the HG y(ka,36) . K min iy, 6. PoE. Kurmyshev, F. Cuevas, R. Sénchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform It is clear that the distance rule assigns an image S to the pattern class S? if'and only if the CCR histogram of S” is the most similar to that ofS, Thus, it is declared that the noisy texture image $”(k,cc,,) belongs to the class S” if the following equation is valid: dy. (8? (ke, )8”)= win bal" be5)5°) (eh2vn2) If the noisy texture image S?(k,ar,) is a (6) signed to the class” , then the recognition is considered to be correct. When the noise level is high enough, the last condition is not satisfied and the noisy texture is not recognizable. 4 Experiments In order to test the technique we applied it to the recognition of P=20 different binary pattern textures S? (p=1,2.....P). Bach pattern texture was computer generated by reproducing a 5x5 random bitmap window (texel) periodically. The image resolution was 100 x 100 pixels Consequently, the generated pattern textures are random locally and periodical in total. A set of six computer generated pattern textures is shown in Figure 2. ‘The test texture database B was generated by adding to each pattern texture S” an uniform noise of different levels k= 0,1,...20, starting from 0 % noise until reaching the 100 °% noisy image. For this purpose we take A=0.05 and = 1,2,.--10 (see Eq. (2)). Thus, in each recognition test for a given noise level K , a set of 200 test texture images is formed by P=20 groups of 10 noisy texture images Sle) Following the scheme represented in Section 3, we then calculate the CCR histogram for each noisy test texture from database B, A 3x3 scanning window was used, so that each HG sy(ba ps) (p= 825205 = 0.20 Gy = 1 QyulO5 B= 12,427) demonstrates the occurrence of histogram 23 =512 possible states of the scanning window. The recognition was accomplished using Eq. (5). Figure 3 shows a set of histograms corresponding to the sample texture 5767 degraded by different level noise. As an example, six different texture images used for classification are shown in Figure 4. In this case, the noise added to each sample texture is 30 % (=6 and A=0.05). These were classified successfully in spite of high-level noise. easaaneaaeal Geseeecseseeeccesaaal ae Figure 2: A subset of six sample textures that were computer generated by using a 5S texel and the binary coded decimal of respective texels, We define the recognition efficiency, for a given noise level K, by the following expression: Ng R=" 100%, © 10P. where 10P is the total number of test textures with noise level K., Nx. is the number of noisy textures with noise level k that were correctly recognized. In terms of the recognition efficiency, the final recognition results are shown in the figure 5. This graph shows the recognition efficiency against noise level. From figure 5 we can see that, in the experiments described above, 100% recognition was achieved in all cases of textures degraded up to 30% by additive uniform noise. In addition, figure 5 shows an interesting phenomenon of recognition revival for textures with the noise level above 60%. The explanation of the phenomenon as follows. First, note that each primary binary texture under the test S” is the periodic translation of a random texel. Second, from the equation (2) we see that the additive uniform noise used in this work changes the binary value of each pixel in accordance 199E. Kurmyshey, F. Cuevas, R. Sanchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform to the noise level k for a given A. A primary image is slightly degraded and well recognized when k <6. Instead, when noise increases, 6 < k < 10 , the image is degraded strongly and recognition decreases to zero. Nevertheless, if noise level is higher, say 10 < < 20, then for the kind of noise used in this work most of pixels have to change their values and some textures are reconstructed partially. The level of reconstruction depends on the regularity of the primary texture. For example, the chessboard image is reconstructed completely. The CCR histogram senses this reconstruction, giving rise to the recognition of the pattern, Note that 100% noise means a negative texture that is not identical to the original positive texture but can have certain morphological similarity to that. “MU tt ; [a cca Noise = 15.86 ° = = BCD ef seaming indow Occurrence Noise #20 ie so [BCD of scanning window Occutrene "000 Noise =26 96 BCD of scanning window Figure 3: A set of five CCR histograms corresponding to the sample texture 5767 degraded by different level noise eS By) be 31668307 Figure 4: Texture i have been correctly ages degraded by 30% noise, those assified by the CCR technique. ‘Recognition efficiency (%4) 100 o o 1029 304050) 390100 Noise Figure 5: Recognition efficiency textures). noise (computer ger In the second experiment, just the same classification technique was evaluated using ten binarized subimages from the Brodatz album (Brodatz, 1966) corrupted by uniform noise. To prepare prototype database we have scanned gray level images D9, D16, D32, D68, D77, D78, D79, D92, D93 and D112. Ten binarized subimages of 100 x 100 pixels, S” (p=, set. They are shown in Fig. 6, A test texture database B is generated by adding to each of ten binary textures S” a uniform noise by Eq, (2). In each recognition test for a given .10), are used in the experiment asthe referenceE. Kurmyshey, F. Cuevas, R. Sanchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform noise level k, a set of 100 test texture images is used. That is formed by P=10 groups of 10 noisy texture images SP (ay), where ce, =12,10. A 3x3 window is used to calculate the CCR histograms. scanning D9 p92 Figure 6: Binarized Brodatz images used to evaluate the CCR technique. Figure 7 shows that the recognition efficiency of Brodatz natural textures versus noise decreases as compared to that of computer generated textures of Fig, 5. The 100% recognition was achieved for all textures degraded up to 5 % by additive noise. This is an expected result because artificial textures of Fig. 2, being periodical, have a narrower statistical distribution in the CCR feature space than that of Brodatz textures. In addition, figure 7 shows the phenomenon of recognition revival for textures with the noise level above 55%, The explanation of the phenomenon is just the same as for the computer-generated textures. Note that images of Fig. 6 are ordered, from jeft to right in line order, to follow from the best recognizable to the worse. That is the D92 has failed first to be recognized at the 10% noise, D9 and D93 failed at the 15% noise, Finally, the DI6 have failed at 40% noise. That means a particular texture can be extremely robust to the uniform noise and can be recognized up to 35% noise. re 7: Recognition efficiency versus noise (binarized Brodatz Recomitian efficiency (%) Noise (* sO textures). 5 Conclusions ‘A method for recognition and classification of binary texture images has been discussed and its application to recognition of texture images degraded by additive uniform noise has been evaluated. One of the available classifiers is used in the CCR feature space in order to assign an image to the category whose prototype it matches best. The CCR histogram is used as the unique attribute for recognition of a binary image. The capacity of the coordinated cluster representation to capture spatial statistical features of an image is exploited. In this paper, the simple minimum distance classifier is used. ‘Two kinds of class prototype images were used, computer generated and natural binary textures. Even though in this work the binary textures, the small 33 scanning window for the coordinated cluster transform and very simple (minimum distance) classification criteria were used in order to test the method for recognition of noise degraded binary images, the performance of the method is highly promising, 100% recognition was achieved for some binary test images degraded by uniform noise up to 30%. As shown in (Kurmyshev and Sénchez-Y éfiez, 2001; Sanchez-Yafiez.et al., 2003; Lépez and Castro, 2002), the combination of the binarization and CCR of textures characterizes the gray level images successfully. Thus, the technique exploited in this work is expected to be adequate for the recognition of real gray level images degraded by noise, for example, in medical imaging where the images used to be noisy Future work should be done to extend the method for gray level and color texture images, to analyze and improve the performance of this recognition method when it is used withE, Kurmyshey, F. Cuevas, R. Sanchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform different size of scanning windows and other classifiers for images degraded by different type of noise Acknowledgments This work was supported by CONACYT under the grant No. 31168-A, Francisco Cuevas thanks the Centro de Investigaciones en Optica, A. C., CONACYT, and Centro de Investigacién en Computacién of the Instituto Politécnico Nacional of México for the support. References Azencott R. and Wang J.P., “Texture classification using windowed Fourier filters”, ZEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 19, No.2, pp.148- 157, 1997, Berry J.R. and Goutsias J., “A Comparison of Matrix Texture Features Using a Maximum Likelihood Texture Classifier” in Visual Communication and Image Processing IV, Proc. SPIE, 1199, pp. 305-316, 1989, Brodatz P., Textures: A Photographie Album for Artists and Designers, Dover Publications, New York, 1966 Chellappa R. and Chatterjee S., “Classification of textures using Gaussian Markov random fields”, IEEE Trans. Acoust Speech Signal Process., Vol. 33, pp. 959-963, 1987. Chen Ch., Pau L.P. and Wang P. S. P., Handbook of Pattern Recognition & Computer Vision, World Scientific, Singapore, 1996, Chetverikoy D., “Texture analysis using feature-based pair wise interaction maps”, Pattern Recognition, Vol. 32, No.3, pp. 487-498, 1999. Duda R.O., Hart P.E. and Stork D.G., “Nonparametric techniques” in Pattern Classification, Wiley-Interscience, John Wiley & Sons Inc., New York, pp. 161-213, 2001 Elfadel LM. and Picard R.W., “Gibbs random fields, cooccurrences and texture modeling”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 16, No.1, pp. 24-31, 1994 Fukunaga K., Introduction to Statistical Pattern Recognition, ‘Academic Press, New York, 1990. Goon A. and Rolland J.P., “Texture classification based on comparison of second-order statistics I: 2P-PDF estimation and distance measure”, .Opt.Soc.Am. A, Vol. 16, No.7, pp. 1566-1574, 1999. Haralick R.M., “Statistical and structural approaches to texture” in Proc. IEEE, Vol. 67, No.5, pp. 786-804, 1979. Kahil M, and Bongard M., Pattern Recognition, Spartan Books, Washington, D.C., 1970. Kurmyshey E.V. and Cervantes M., “A quasi-statistical approach to digital image representation”, Rev, Mex. Fis., Vol. 42, No.1, pp. 104-116, 1996 Kurmyshey E.V. and Soto R., “Digital Pattern Recognition in the coordinated cluster representation” in NORSIG 96 (1996 IEEE NORDIC SIGNAL PROCESSING SYMPOSIUM), Espoo, Finland, pp. 463-466, 1996. Kurmyshey E.V. and Sanchez-Yaiez R.E., “Texture Classification based on Image representation by Coordinated Clusters” in Proceedings Image and Vision Computing New Zealand 2001, Dunedin, NZ, pp. 213-217, 2001 Liu S. Sh. and Jernigan M.E., “Texture analysis and discrimination in additive noise”, Computer Vision, Graphics and Image Processing, Vol. 49, No.1, pp. 52-69, 1990. Lopez, D. and Castro, M.J., “Neural-based classification of blocks from documents”, International Conference on Neural Networks 2002, Madrid, Espatia. Ohanian P,P, and Dubes R.C. four classes of textural feature 25, No. 8, pp. 819-833, 1992. “Performance evaluation for ', Pattern Recognition, Vol Ojala T., Pietikainen M. and Harwood D.,““A comparative study of texture measures with classification based on feature distributions”, Pattern Recognition, Vol. 29, No.1, pp. 51-59, 1996, Randen T. and Husoy J.H., “Filtering for Texture Classification: A Comparative Study”, JEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 21, No. 4, 1999, pp. 291-310. Sanchez-Yafiez R.E., Kurmyshey E.V, and Cuevas F.J.,°A Framework for Texture Classification using the Coordinated Clusters Representation”, Pattern Recognition Letters, Vol. 24, pp. 21-31, 2003, 1s Fd “A new kind of neural network functions”, Rev. Mex. Fis., Vol. 39, No.2, pp. 235-249, 1993, Soh L.K. and Tsatsoulis C., “Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices”, IEEE Transactions on Geosciences and Remote Sensing, Vol. 37, No.2, pp. 780-787, 1999.E, Kurmyshey, F. Cuevas, R. Sanchez: Noisy Binary Texture Recognition Using the Coordinated Cluster Transform Thomson M.G.A. and Foster D.H., “Role of second- and Valkealahti K. and Oja E., “Reduced multidimensional co- third-order statistics in the discriminability of natural images", occurrence histograms in texture analysis”, JEEE LOpt.Soc.Am. A, Vol. 14, No.9, pp. 2081-2090, 1997, Transactions on Pattern Analysis and Machine Intelligence, Vol. 20, No. 1, pp. 90-95, 1998. ‘Tuceryan M. and Jain A.K. “Texture Analysis” in Handbook of Pattern Recognition and Computer Vision, (CH. Chen, — Wang, L. and He D.C., “Texture classification using texture LF. Pau and P.S.P Wang Eds.), World Scientific Publishing spectrum”, Pattern Recognition, Vol. 23, pp. 905-910, 1990, Company, Singapore, pp. 235-276, 1993. ‘Turner MLR., “Texture discrimination by Gabor functions”, Biol. Cyber. Vol. 55, pp. 71-82, 1986. Evguenii Kourmychev (Kurmyshev), received his Master degree in Theoretical and Mathematical Physics from the Moscow Physical Technical Institute, Russia in 1975 and his Ph.D. in Theoretical and Mathematical Physics from PN. Lebedev Physics Institute of the Academy of Sciences ofthe USSR in 1982. Since 1991 he resides in Mexico and currently holds a research position (bwvestigador Titular) at the Centro de Investigaciones en Optica, A.C. in Leén, Mexico and aprofessor ofpasigraduate programs atthe same Center: He isa fellow of National System of Researchers of Mexico (SNJ) and the author of more than sixty publications and two textbooks. His research interests include but are not limited to digital texture image analysis and expert computer vision systems. Francisco Javier Cuevas de ta Rosa, received his BS degree in Computer Systems Engineering from Instituto Teenolégico y de Estudios Superiores de Monterrey (ITESM) Campus Monterrey in 1984. He obtained his Master degree in Computer Science from ITESM Campus Edo. de México in 1995 and his PhD in Opties Metrology from Centro de Investigaciones en Optica, A.C. in 2000, where he is currently a researcher. He made a posdoctoral stay in the Centro de Investigacién en Computacion of the Instituto Politéenico Nacional of México, between 2001 and 2002. He has more than 20 publications in international journals with rigorous refereeing and more than 40 works in national and international conferences. He is a member of the Sistema Nacional de Investigadores (SNI) of México since 1995. His research interests spread over Optical Metrology. Computer Vision, Digital Image Processing, Pattern recognition, Neural networks and Genetic Algorithms. Rati! E, Sanchez Yanez, received Master degree in Electric Engineering from the Universidad de Guanajuato, México in 1995 and his Ph.D. in Sciences (Optics) from the Centro de Investigaciones en Optica, A.C., Ledn, México in 2002, The dissertation was done in the field of texture image analysis and classification. His professional interests are digital image analysis and processing, intelligent systems (particularly. fuzzy logic and artificial neural networks), computer systems and education. The main direction of his work is the implementation of computer integrated environment in C language, He gave various lecture courses for graduate students. Since 1991 he és a fellow of The Institute of lectrical and Electronics Engineers (IBEE) and IEEE Computer Society and IEEE Systems ‘Man and Cybernetics Society since 1995. 203